A practical guide to developments in data imputation methods

This paper provides a brief guide to recent developments in data imputation methods which are of value in transport studies. 'Non-response' or 'missing data' are found in many types of surveys commonly used in transportation research and development control practice. A number of the more common (imputation) methods, used to 'fill the gaps' when non-response has occurred, are demonstrated using a sample dataset taken from the Trip Rate Information Computer System (TRICS). In addition, some of the more recent and less well known developments such as the Approximate Bayesian Bootstrap (ABB) are discussed. Based upon a simple trip generation model (for a selection of UK Office developments), it is shown that the alternative methods can potentially lead to serious 'missing data bias'. Accordingly, it is suggested that considerable caution is exercised when drawing policy implications out of data analysis featuring imputed elements. A transparency principle is suggested as an appropriate guiding tool when dealing with transport survey data with missing elements.